A two-stage architecture for identifying and locating the source of pain using novel multi-domain binary patterns of EDA

Pain, an extremely unpleasant sensory experience, lacks an objective diagnostic test for accurate measurement. When individuals are unable to communicate, identifying and locating pain becomes crucial for improving treatment outcomes. Despite numerous studies on pain identification, a reliable consensus has yet to be reached. This study, utilising the AI4Pain dataset, aims to establish a strong correlation between Electrodermal Activity (EDA) signal features and the presence of acute pain, as well as clarify the relationship between classified signals and the pain's location. To this end, EDA signals were recorded from 61 subjects while inducing electrical pain in either of two anatomical locations (hand and forearm) for each subject. The EDA data underwent preprocessing to eliminate irrelevant information using a Butterworth IIR bandpass filter and a median filter. A novel feature descriptor called Multi-Domain Binary Patterns (MDBP) was proposed for this research. These MDBPs were combined with time domain features, and a reduced feature vector was obtained using Minimum Redundancy Maximum Relevance (MRMR). The resulting vector then formed the input of ensemble classification algorithms. The proposed method consists of two stages: The first stage focuses on pain detection, while the second stage focuses on pain localisation. Using leave-one-subject-out cross-validation, the proposed method achieved an accuracy of 77.9% in pain detection (Stage I), while the pain localisation experiment (Stage II) resulted in an accuracy of 69.67%. The efficacy of the proposed method was also validated through the publicly available BioVid database.